387-page free LangChain 101 manual digs into source code, finds framework doesn't boost accuracy
ghumare64 · x · 2026-11-20
The author released a free, open-source 387-page "LangChain 101" manual covering Agents, Graphs, Memory and Tools. Unlike typical tutorials, it reads the source of LangChain 1.5, LangGraph 1.2, Deep Agents 0.7 and LangSmith, with reproducible experiments on agent crashes, retries, state loss and resumption.
Key findings:
- Across 60 shared local benchmark tasks, a plain tool loop, createagent, and a hand-built StateGraph produced identical outcomes — LangChain doesn't improve agent accuracy;
- Deep Agents carries a real token cost, with surprising prompt-token results;
- Checkpointing doesn't guarantee exactly-once execution.
More from coding & agent
- How a 4,888-card Obsidian Kanban board went from 9 seconds to under 3 seconds — dSebastien · 2026-11-20
- Dev hails sign-in with Claude as the week's best launch for offloading pricing complexity — threepointone · 2026-11-20
- Zero-security dev's autonomous agent went from one-byte overflow to tcache poisoning in hours — Genaforvena · 2026-11-20
- Rhenium: a from-scratch Rust decompiler built for humans and AI agents — OwariDa · 2026-11-20
- RUC & Fudan release CoworkOS: agent runtime for long-horizon work with human edits, boosting benchmarks from 33.19 to 42.95 — jiqizhixin · 2026-11-20
- Starlight Arena offers $50k in prizes for talking AI agents into breaking their own rules — jim-jam-biscuit · 2026-11-20